Critical Values and Significance Testing
What You'll Learn
You'll learn how to use critical value tables to decide whether your calculated Grubbs' test statistic indicates a truly significant outlier or just normal variation.
The Decision Framework
Think of critical values as a courtroom threshold: your test statistic is the evidence, and the critical value is the standard of proof needed to convict a data point of being an outlier.
After calculating your Grubbs' test statistic (which measures how extreme a suspected outlier is), you need to compare it against a critical value to make your final call. The critical value depends on two things:
- Your significance level (α): Typically 0.05 (5% chance of false alarm) or 0.01 (1%)
- Your sample size (n): Larger datasets have different thresholds
How the Comparison Works
If your test statistic > critical value: The data point is statistically significant—it's likely a genuine outlier, not random chance.
If your test statistic ≤ critical value: Insufficient evidence; treat it as normal variation.
You find critical values in Grubbs' test tables (organized by sample size and α level) or calculate them using formulas involving the t-distribution. For example, with n=30 and α=0.05, the critical value might be 2.745. If your test statistic is 3.12, you'd reject the null hypothesis and flag that point as an outlier.
Why This Matters
Without this step, you're just calculating numbers without context. The critical value transforms your statistic into an actionable decision, balancing the risk of false positives (flagging normal data) against missing real anomalies.
Key Takeaway: Critical values act as decision thresholds—only when your test statistic exceeds them can you confidently declare a data point a statistically significant outlier at your chosen confidence level.